From a829b8f3211c2f5b0851cdd7e0f2417d9ab06f51 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Petr=20Balv=C3=ADn?= <opensource@petrbalvin.org>
Date: Wed, 7 Oct 2026 02:09:46 +0200
Subject: [PATCH] fix(parser): read the segment-absolute rendering FS:0

The toolchain's disassembler prints the segment-prefixed disp32
absolute as FS:0, but the bare-name branch read the segment register
alone and dropped the offset, so MOVQ FS:0, DX silently encoded a
register move.  The colon-offset form now lowers to the same
segment-absolute operand the 0(FS) spelling takes.

Assisted-by: GLM 5.3 Flash
---
 asm/rendered_text_test.go | 48 +++++++++++++++++++++++++++++++++++++++
 parser/parser.go          | 15 ++++++++++++
 2 files changed, 63 insertions(+)
 create mode 100644 asm/rendered_text_test.go

diff --git a/asm/rendered_text_test.go b/asm/rendered_text_test.go
new file mode 100644
index 0000000..e3fe9e1
--- /dev/null
+++ b/asm/rendered_text_test.go
@@ -0,0 +1,48 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+package asm
+
+import (
+	"encoding/hex"
+	"strings"
+	"testing"
+
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
+)
+
+// TestRenderedTextEncodes pins the encoder against the disassembler's
+// renderings: every line here is a text the toolchain's own decoder prints
+// for bytes its assembler produced (the disasm package's parity fixtures),
+// so assembling the printed text must reproduce those bytes.  Each row was
+// read back through the toolchain's decoder; the hex is the fixture's.
+func TestRenderedTextEncodes(t *testing.T) {
+	for _, tt := range []struct {
+		text string
+		want string
+	}{
+		// The segment-absolute rendering lowers to the FS-prefixed disp32
+		// absolute, exactly what the 0(FS) spelling encodes.
+		{"MOVQ FS:0, DX", "64488b142500000000"},
+	} {
+		src := "TEXT \u00b7k(SB), NOSPLIT, $0\n\t" + tt.text + "\n\tRET\n"
+		f, errs := parser.Parse("k.s", src)
+		if len(errs) > 0 {
+			t.Errorf("%s: parse: %v", tt.text, errs[0])
+			continue
+		}
+		img, err := AssembleFile(f)
+		if err != nil {
+			t.Errorf("%s: assemble: %v", tt.text, err)
+			continue
+		}
+		fn := img.Funcs[0]
+		code := img.Code[fn.Offset : fn.Offset+fn.Size]
+		// The no-fallthrough rule appends the RET; the pinned instruction is
+		// the prefix.
+		got := hex.EncodeToString(code)
+		if !strings.HasPrefix(got, tt.want) {
+			t.Errorf("%s: bytes %s, want prefix %s", tt.text, got, tt.want)
+		}
+	}
+}
diff --git a/parser/parser.go b/parser/parser.go
index f0e4c38..c49e629 100644
--- a/parser/parser.go
+++ b/parser/parser.go
@@ -678,6 +678,21 @@ func parseAddress(g []token.Token) ast.Address {
 			i++
 		}
 	}
+	// Segment-absolute in the rendering spelling: FS:0 and GS:0.  The
+	// colon-offset form is what the toolchain's disassembler prints for the
+	// segment-prefixed disp32 absolute, so the rendered text must re-encode;
+	// without it the segment register alone parsed and the offset vanished.
+	// The base field naming the segment lowers to the same operand the
+	// 0(FS) spelling takes.
+	if len(g) >= 3 && g[0].Kind == token.Ident && g[1].Kind == token.Colon &&
+		(g[0].Text == "FS" || g[0].Text == "GS") {
+		if v, rest, ok := foldExpr(g[2:]); ok && len(rest) == 0 {
+			addr.Base = g[0].Text
+			addr.Offset = v
+			addr.HasOff = true
+			return addr
+		}
+	}
 	// Bare name (register, label or symbol) possibly with an arm64 shift.
 	if addr.Base == "" && addr.Sym == nil && g[0].Kind == token.Ident {
 		sym := &ast.Symbol{Pos: g[0].Pos}